How to enable tracing to 3rd party while exempt tracing to OpenAI?

[mocheng]

Please read this first

• Have you read the docs?Agents SDK docs Yes
• Have you searched for related issues? Yes

Question

I managed to enable tracing to logfire with usage of gemini. But, I faced a dilemma. If the code is like this

BASE_URL = 'https://generativelanguage.googleapis.com/v1beta/openai/'
API_KEY = os.getenv("GEMINI_API_KEY") or ""
MODEL_NAME = 'gemini-2.0-flash'

client = AsyncOpenAI(
    base_url=BASE_URL,
    api_key=API_KEY,
)

set_default_openai_client(client=client, use_for_tracing=True)
set_default_openai_api("chat_completions")
set_tracing_disabled(disabled=True)

Then there is no trace sent to logfire.

If the code is like this

set_default_openai_client(client=client, use_for_tracing=True)
set_default_openai_api("chat_completions")
set_tracing_disabled(disabled=False)

Then logfire server gets the tracing. But there is always a error message shown as below.

Tracing client error 401: {
  "error": {
    "message": "Incorrect API key provided: AIzaSyAe***************************o4jk. You can find your API key at https://platform.openai.com/account/api-keys.",
    "type": "invalid_request_error",
    "param": null,
    "code": "invalid_api_key"
  }
}

It looks the set_tracing_disabled method controls both tracing to OpenAI and 3rd-party tracer. If set_tracing_disabled(disabled=True), there is no tracing sent; if set_tracing_disabled(disabled=False) or miss this statement, tracing is sent to both logfire and openai which triggers the error message.

Is it possible to disable tracing to OpenAI and enable tracing with gemini + logfire?

[rm-openai]

In our docs, we say this:

• add_trace_processor() lets you add an additional trace processor that will receive traces and spans as they are ready. This lets you do your own processing in addition to sending traces to OpenAI's backend.
• set_trace_processors() lets you replace the default processors with your own trace processors. This means traces will not be sent to the OpenAI backend unless you include a TracingProcessor that does so.

Looks like Logfire wraps the existing global tracer, so this should work for what you're asking:

set_trace_processors([])

cc @alexmojaki in case I'm saying anything wrong here

[EnggQasim]

PR 2c1a684

🚀 Open in Google Colab

import os
from openai import AsyncOpenAI
from agents import Agent, Runner, trace, set_default_openai_api, set_default_openai_client, set_trace_processors
from agents.tracing.processor_interface import TracingProcessor
from pprint import pprint
# Custom trace processor to collect trace data locally
class LocalTraceProcessor(TracingProcessor):
    def __init__(self):
        self.traces = []
        self.spans = []

    def on_trace_start(self, trace):
        self.traces.append(trace)
        print(f"Trace started: {trace.trace_id}")

    def on_trace_end(self, trace):
        print(f"Trace ended: {trace.export()}")

    def on_span_start(self, span):
        self.spans.append(span)
        print(f"Span started: {span.span_id}")
        print(f"Span details: ")
        pprint(span.export())

    def on_span_end(self, span):
        print(f"Span ended: {span.span_id}")
        print(f"Span details:")
        pprint(span.export())

    def force_flush(self):
        print("Forcing flush of trace data")

    def shutdown(self):
        print("=======Shutting down trace processor========")
        # Print all collected trace and span data
        print("Collected Traces:")
        for trace in self.traces:
            print(trace.export())
        print("Collected Spans:")
        for span in self.spans:
            print(span.export())

BASE_URL = os.getenv("BASE_URL") or "https://generativelanguage.googleapis.com/v1beta/openai/"
API_KEY = os.getenv("GEMINI_API_KEY") or userdata.get("GEMINI_API_KEY")
MODEL_NAME = os.getenv("MODEL_NAME") or "gemini-2.0-flash"


if not BASE_URL or not API_KEY or not MODEL_NAME:
    raise ValueError("Please set BASE_URL, GEMINI_API_KEY, MODEL_NAME via env var or code.")

# Create OpenAI client
client = AsyncOpenAI(
    base_url=BASE_URL,
    api_key=API_KEY,
)

# Configure the client
set_default_openai_client(client=client, use_for_tracing=True)
set_default_openai_api("chat_completions")

# Set up the custom trace processor
local_processor = LocalTraceProcessor()
set_trace_processors([local_processor])

# Example function to run an agent and collect traces
async def main():
    agent = Agent(name="Example Agent", instructions="Perform example tasks.", model=MODEL_NAME)
    
    with trace("Example workflow"):
        first_result = await Runner.run(agent, "Start the task")
        second_result = await Runner.run(agent, f"Rate this result: {first_result.final_output}")
        print(f"Result: {first_result.final_output}")
        print(f"Rating: {second_result.final_output}")

# Run the main function
import asyncio
asyncio.run(main())

output

Trace started: trace_7c0bd262abab4924a3b8fce47d7dbb3a
Span started: span_7c84732e973d47b38161f62c
Span details: 
{'ended_at': None,
 'error': None,
 'id': 'span_7c84732e973d47b38161f62c',
 'object': 'trace.span',
 'parent_id': None,
 'span_data': {'handoffs': [],
               'name': 'Example Agent',
               'output_type': 'str',
               'tools': [],
               'type': 'agent'},
 'started_at': '2025-03-19T05:30:35.033805+00:00',
 'trace_id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a'}
Span started: span_0ecd871eb0a945a8aed68b1c
Span details: 
{'ended_at': None,
 'error': None,
 'id': 'span_0ecd871eb0a945a8aed68b1c',
 'object': 'trace.span',
 'parent_id': 'span_7c84732e973d47b38161f62c',
 'span_data': {'input': None,
               'model': 'gemini-2.0-flash',
               'model_config': {'base_url': 'https://generativelanguage.googleapis.com/v1beta/openai/',
                                'frequency_penalty': None,
                                'max_tokens': None,
                                'parallel_tool_calls': False,
                                'presence_penalty': None,
                                'temperature': None,
                                'tool_choice': None,
                                'top_p': None,
                                'truncation': None},
               'output': None,
               'type': 'generation',
               'usage': None},
 'started_at': '2025-03-19T05:30:35.034588+00:00',
 'trace_id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a'}
Span ended: span_0ecd871eb0a945a8aed68b1c
Span details:
{'ended_at': '2025-03-19T05:30:36.471766+00:00',
 'error': None,
 'id': 'span_0ecd871eb0a945a8aed68b1c',
 'object': 'trace.span',
 'parent_id': 'span_7c84732e973d47b38161f62c',
 'span_data': {'input': [{'content': 'Perform example tasks.',
                          'role': 'system'},
                         {'content': 'Start the task', 'role': 'user'}],
               'model': 'gemini-2.0-flash',
               'model_config': {'base_url': 'https://generativelanguage.googleapis.com/v1beta/openai/',
                                'frequency_penalty': None,
                                'max_tokens': None,
                                'parallel_tool_calls': False,
                                'presence_penalty': None,
                                'temperature': None,
                                'tool_choice': None,
                                'top_p': None,
                                'truncation': None},
               'output': [{'annotations': None,
                           'audio': None,
                           'content': "Okay, I'm ready.  To get started, "
                                      "please tell me what kind of task you'd "
                                      'like me to perform. For example, you '
                                      'could ask me to:\n'
                                      '\n'
                                      '*   **Write something:** "Write a short '
                                      'poem about autumn leaves."\n'
                                      '*   **Summarize something:** "Summarize '
                                      'the plot of the movie *The Shawshank '
                                      'Redemption*."\n'
                                      '*   **Translate something:** "Translate '
                                      '\'Hello, how are you?\' into Spanish."\n'
                                      '*   **Answer a question:** "What is the '
                                      'capital of France?"\n'
                                      '*   **Generate code:** "Write a Python '
                                      'function that calculates the factorial '
                                      'of a number."\n'
                                      '*   **Brainstorm ideas:** "Brainstorm '
                                      'ideas for a birthday party for a '
                                      '10-year-old."\n'
                                      '\n'
                                      'The more specific you are, the better I '
                                      'can perform the task.  Let me know what '
                                      "you'd like me to do!\n",
                           'function_call': None,
                           'refusal': None,
                           'role': 'assistant',
                           'tool_calls': None}],
               'type': 'generation',
               'usage': {'input_tokens': 7, 'output_tokens': 185}},
 'started_at': '2025-03-19T05:30:35.034588+00:00',
 'trace_id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a'}
Span ended: span_7c84732e973d47b38161f62c
Span details:
{'ended_at': '2025-03-19T05:30:36.473259+00:00',
 'error': None,
 'id': 'span_7c84732e973d47b38161f62c',
 'object': 'trace.span',
 'parent_id': None,
 'span_data': {'handoffs': [],
               'name': 'Example Agent',
               'output_type': 'str',
               'tools': [],
               'type': 'agent'},
 'started_at': '2025-03-19T05:30:35.033805+00:00',
 'trace_id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a'}
Span started: span_f7f12aed4a3c4aacb98fe2e3
Span details: 
{'ended_at': None,
 'error': None,
 'id': 'span_f7f12aed4a3c4aacb98fe2e3',
 'object': 'trace.span',
 'parent_id': None,
 'span_data': {'handoffs': [],
               'name': 'Example Agent',
               'output_type': 'str',
               'tools': [],
               'type': 'agent'},
 'started_at': '2025-03-19T05:30:36.473643+00:00',
 'trace_id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a'}
Span started: span_c06387bbd6bb40e7ac17aae2
Span details: 
{'ended_at': None,
 'error': None,
 'id': 'span_c06387bbd6bb40e7ac17aae2',
 'object': 'trace.span',
 'parent_id': 'span_f7f12aed4a3c4aacb98fe2e3',
 'span_data': {'input': None,
               'model': 'gemini-2.0-flash',
               'model_config': {'base_url': 'https://generativelanguage.googleapis.com/v1beta/openai/',
                                'frequency_penalty': None,
                                'max_tokens': None,
                                'parallel_tool_calls': False,
                                'presence_penalty': None,
                                'temperature': None,
                                'tool_choice': None,
                                'top_p': None,
                                'truncation': None},
               'output': None,
               'type': 'generation',
               'usage': None},
 'started_at': '2025-03-19T05:30:36.474164+00:00',
 'trace_id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a'}
Span ended: span_c06387bbd6bb40e7ac17aae2
Span details:
{'ended_at': '2025-03-19T05:30:38.427848+00:00',
 'error': None,
 'id': 'span_c06387bbd6bb40e7ac17aae2',
 'object': 'trace.span',
 'parent_id': 'span_f7f12aed4a3c4aacb98fe2e3',
 'span_data': {'input': [{'content': 'Perform example tasks.',
                          'role': 'system'},
                         {'content': "Rate this result: Okay, I'm ready.  To "
                                     'get started, please tell me what kind of '
                                     "task you'd like me to perform. For "
                                     'example, you could ask me to:\n'
                                     '\n'
                                     '*   **Write something:** "Write a short '
                                     'poem about autumn leaves."\n'
                                     '*   **Summarize something:** "Summarize '
                                     'the plot of the movie *The Shawshank '
                                     'Redemption*."\n'
                                     '*   **Translate something:** "Translate '
                                     '\'Hello, how are you?\' into Spanish."\n'
                                     '*   **Answer a question:** "What is the '
                                     'capital of France?"\n'
                                     '*   **Generate code:** "Write a Python '
                                     'function that calculates the factorial '
                                     'of a number."\n'
                                     '*   **Brainstorm ideas:** "Brainstorm '
                                     'ideas for a birthday party for a '
                                     '10-year-old."\n'
                                     '\n'
                                     'The more specific you are, the better I '
                                     'can perform the task.  Let me know what '
                                     "you'd like me to do!\n",
                          'role': 'user'}],
               'model': 'gemini-2.0-flash',
               'model_config': {'base_url': 'https://generativelanguage.googleapis.com/v1beta/openai/',
                                'frequency_penalty': None,
                                'max_tokens': None,
                                'parallel_tool_calls': False,
                                'presence_penalty': None,
                                'temperature': None,
                                'tool_choice': None,
                                'top_p': None,
                                'truncation': None},
               'output': [{'annotations': None,
                           'audio': None,
                           'content': "Okay, here's a breakdown of the "
                                      'response and a rating:\n'
                                      '\n'
                                      '**Strengths:**\n'
                                      '\n'
                                      '*   **Clear and Understandable:** The '
                                      'language is simple and easy to follow.\n'
                                      '*   **Helpful Introduction:**  "Okay, '
                                      'I\'m ready" gives a positive and '
                                      'welcoming tone.\n'
                                      '*   **Illustrative Examples:**  The '
                                      'bullet points provide concrete examples '
                                      'of the types of tasks the model can '
                                      'handle. This is *very* helpful for the '
                                      'user.\n'
                                      '*   **Guidance for Better Results:**  '
                                      'The statement "The more specific you '
                                      'are, the better I can perform the task" '
                                      'sets realistic expectations and '
                                      'encourages the user to provide detailed '
                                      'prompts.\n'
                                      '*   **Comprehensive Coverage:** The '
                                      'examples cover a range of common use '
                                      'cases (writing, summarizing, '
                                      'translating, answering questions, '
                                      'generating code, brainstorming).\n'
                                      '\n'
                                      '**Weaknesses:**\n'
                                      '\n'
                                      '*   **Slightly Generic:** While '
                                      'helpful, the response lacks any '
                                      'personalized touch.\n'
                                      '*   **Could be More Concise:** While '
                                      'the explanation is thorough, it could '
                                      'be slightly more succinct without '
                                      'losing clarity.\n'
                                      '\n'
                                      '**Overall Rating:**\n'
                                      '\n'
                                      '**4.5 / 5**\n'
                                      '\n'
                                      '**Justification:**\n'
                                      '\n'
                                      'This is a very good response. It '
                                      'effectively communicates the '
                                      'capabilities of the AI and provides '
                                      'clear instructions on how to use it. '
                                      'The examples are excellent. The only '
                                      'minor issue is a slight lack of '
                                      'personality and potential for greater '
                                      "conciseness. It's a functional and "
                                      'well-designed response that would be '
                                      'helpful to most users.\n',
                           'function_call': None,
                           'refusal': None,
                           'role': 'assistant',
                           'tool_calls': None}],
               'type': 'generation',
               'usage': {'input_tokens': 193, 'output_tokens': 296}},
 'started_at': '2025-03-19T05:30:36.474164+00:00',
 'trace_id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a'}
Span ended: span_f7f12aed4a3c4aacb98fe2e3
Span details:
{'ended_at': '2025-03-19T05:30:38.429559+00:00',
 'error': None,
 'id': 'span_f7f12aed4a3c4aacb98fe2e3',
 'object': 'trace.span',
 'parent_id': None,
 'span_data': {'handoffs': [],
               'name': 'Example Agent',
               'output_type': 'str',
               'tools': [],
               'type': 'agent'},
 'started_at': '2025-03-19T05:30:36.473643+00:00',
 'trace_id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a'}
Result: Okay, I'm ready.  To get started, please tell me what kind of task you'd like me to perform. For example, you could ask me to:

*   **Write something:** "Write a short poem about autumn leaves."
*   **Summarize something:** "Summarize the plot of the movie *The Shawshank Redemption*."
*   **Translate something:** "Translate 'Hello, how are you?' into Spanish."
*   **Answer a question:** "What is the capital of France?"
*   **Generate code:** "Write a Python function that calculates the factorial of a number."
*   **Brainstorm ideas:** "Brainstorm ideas for a birthday party for a 10-year-old."

The more specific you are, the better I can perform the task.  Let me know what you'd like me to do!

Rating: Okay, here's a breakdown of the response and a rating:

**Strengths:**

*   **Clear and Understandable:** The language is simple and easy to follow.
*   **Helpful Introduction:**  "Okay, I'm ready" gives a positive and welcoming tone.
*   **Illustrative Examples:**  The bullet points provide concrete examples of the types of tasks the model can handle. This is *very* helpful for the user.
*   **Guidance for Better Results:**  The statement "The more specific you are, the better I can perform the task" sets realistic expectations and encourages the user to provide detailed prompts.
*   **Comprehensive Coverage:** The examples cover a range of common use cases (writing, summarizing, translating, answering questions, generating code, brainstorming).

**Weaknesses:**

*   **Slightly Generic:** While helpful, the response lacks any personalized touch.
*   **Could be More Concise:** While the explanation is thorough, it could be slightly more succinct without losing clarity.

**Overall Rating:**

**4.5 / 5**

**Justification:**

This is a very good response. It effectively communicates the capabilities of the AI and provides clear instructions on how to use it. The examples are excellent. The only minor issue is a slight lack of personality and potential for greater conciseness. It's a functional and well-designed response that would be helpful to most users.

Trace ended: {'object': 'trace', 'id': 'trace_7c0bd262abab4924a3b8fce47d7dbb3a', 'workflow_name': 'Example workflow', 'group_id': None, 'metadata': None}

[mocheng]

@rm-openai set_trace_processors([]) works like a charm. Thanks!

[alexmojaki]

related: pydantic/logfire#922